joseph.ergo@proton.me | Portfolio | Resume PDF | Linked-In | +212 713-617-633

Available immediately for full/part-time remote roles

MU SIGMA INC.

Table of contents

• 1 About the project
• 2 EDA
  • 2.1 information technology and services indestry’s yearly new recruit count
  • 2.2 mu sigma inc.’s workforce status over the years
• 3 Persona company network graph
• 4 Workforce sample
  • 4.1 Abhijith Asok
  • 4.2 Anish Mahapatra
  • 4.3 Archisman Dinda
  • 4.4 Ashish Kumar
  • 4.5 Ashwin Avasarala
  • 4.6 Aviral Agrawal
  • 4.7 Balaji Soundararajan
  • 4.8 Bryce Doreen
  • 4.9 Chinedu Ndukwe
  • 4.10 Isha Vyas
  • 4.11 Jayadev Chakravarty
  • 4.12 Julie Woods-Moss
  • 4.13 Mark Dzialga
  • 4.14 Neeratyoy Mallik
  • 4.15 Noreen Madden
  • 4.16 Pankhuri Vijay
  • 4.17 Piyush Mundhra
  • 4.18 Praveen Suresh
  • 4.19 Raghavendra Ramesh
  • 4.20 Rahul Kishore
  • 4.21 Rajat Sharma
  • 4.22 Ramon Serrallonga
  • 4.23 Rituparna Roy
  • 4.24 Saul Bravo
  • 4.25 Sreekanth Mahesala
  • 4.26 Stephen Wu
  • 4.27 Sujit Menon
  • 4.28 Vinayakam Murugan
  • 4.29 Vivek Jha
  • 4.30 Yuedong Lu

## SETUP
from pathlib import Path
import duckdb
from tqdm.notebook import tqdm
import datetime
import copy
import polars as pl
import plotly.express as px
import plotly.io as pio
import re
from concurrent.futures import ThreadPoolExecutor
import plotly.graph_objects as go
import networkx as nx
import numpy as np
# pio.renderers.default = 'plotly_mimetype'
pio.renderers.default = 'jupyterlab+notebook'
pio.templates.default = "plotly_white"

path_data = Path.cwd()/'data'/'03_rdb'
path_data_companies = path_data/'companies_table.parquet'
path_data_experience = path_data/'experience_table.parquet'
path_data_emails = path_data/'emails_table.parquet'
path_data_education = path_data/'education_table.parquet'
path_data_school = path_data/'school_table.parquet'
path_data_persona = path_data/'persona_table.parquet'
path_data_profiles = path_data/'profiles_table.parquet'

path_output_images = Path.cwd()/'output'/'images'

conn = duckdb.connect()

conn.execute("SET temp_directory = 'temp';")
conn.execute("SET memory_limit = '10GB';")
conn.execute("SET max_temp_directory_size = '100GB';")
conn.execute("SET threads = 8;")
conn.execute("SET preserve_insertion_order = false;")
conn.execute("SET enable_progress_bar = true;")
conn.execute("SET enable_progress_bar_print = true;")

df = pl.read_parquet('03_target_companies3.parquet')
df_yearly_new_hires_per_indestry = pl.read_parquet('03_yearly_new_hires_per_indestry.parquet')

current_company_id = "&-friends"
current_company_id = pl.read_json("04__control__.json")[0,'current_company_id']
query = f"""
SELECT *
FROM read_parquet('{path_data_companies}')
WHERE company_id = '{current_company_id}'
"""
df_company_by_company_id = pl.DataFrame(conn.execute(query).df())

current_company_name = df_company_by_company_id[0,'company_name']
current_company_indestry = df_company_by_company_id[0,'company_industry']

current_company_parquet = Path.cwd()/'output'/'company_data'/f"{current_company_id}.parquet"

# Info about personas status from company_id

query = f"""
SELECT *
FROM read_parquet('{path_data_experience}')
WHERE company_id = '{current_company_id}'
"""
df_experiences_by_company_id = pl.DataFrame(conn.execute(query).df())

personas_whitout_end_date = df_experiences_by_company_id.filter(pl.col('end_date').is_null())
personas_who_got_raise = df_experiences_by_company_id.filter((pl.col('end_date').is_not_null()) &
                                     pl.col('persona_id').is_in(personas_whitout_end_date['persona_id'].to_list()))
personas_who_stayed = (pl
                      .concat([personas_whitout_end_date, personas_who_got_raise])
                      .sort('start_date')
                      .group_by('persona_id')
                      .agg(
                          pl.col('title_name').last(),
                          pl.col('is_primary').last(),
                          pl.col('start_date').min(),
                          pl.col('end_date').max(),
                          pl.col('title_name').count().alias('changes'),
                          pl.col('title_name').unique().alias('all_title_name'),
                      )
                      .with_columns(
                          pl.lit(True).alias('still_associated'),
                          pl.lit(None).alias('end_date')
                      )
                      .sort('changes')
                             )

personas_who_left = df_experiences_by_company_id.filter((pl.col('end_date').is_not_null()) & ~pl.col('persona_id').is_in(personas_who_stayed['persona_id'].to_list()) )
personas_who_left = (personas_who_left
                     .sort('start_date')
                     .group_by('persona_id')
                     .agg(
                          pl.col('title_name').last(),
                          pl.col('is_primary').last(),
                          pl.col('start_date').min(),
                          pl.col('end_date').max(),
                          pl.col('title_name').count().alias('changes'),
                          pl.col('title_name').unique().alias('all_title_name'),
                              )
                     .with_columns(
                         pl.lit(False).alias('still_associated'),
                         
                     )
                     .sort('changes'))

df_personas_who_worked_in_company = pl.concat([personas_who_stayed, personas_who_left], how='vertical_relaxed').with_columns(
    (pl.col('end_date').dt.year()-pl.col('start_date').dt.year()).alias('work_durration')
).sort('work_durration')

import dns.resolver
import smtplib
import socket

def check_deliverability(email_address):
    """
    Checks the deliverability of an email address by verifying MX records
    and performing an SMTP connection test.
    """
    if '@' not in email_address:
        return False
    
    domain = email_address.split('@')[1]
    
    # Check for MX records
    try:
        mx_records = dns.resolver.resolve(domain, 'MX')
        if not mx_records:
            return False
    except (dns.resolver.NoAnswer, dns.resolver.NXDOMAIN, dns.resolver.Timeout):
        return False

    # Perform SMTP connection test
    mx_host = str(mx_records[0].exchange)
    
    # Validate MX hostname before attempting connection
    try:
        # Test if hostname can be properly encoded
        mx_host.encode('idna')
    except UnicodeError:
        return False
    
    try:
        with smtplib.SMTP(mx_host, timeout=10) as smtp:
            smtp.set_debuglevel(0)
            smtp.helo(socket.gethostname())
            smtp.mail('test@example.com')
            code, _ = smtp.rcpt(email_address)

            return code == 250  # 250 indicates valid email address
            
    except (smtplib.SMTPException, socket.error, UnicodeError):
        return False

# info of all personas info
list_w = []
for word in df_experiences_by_company_id['persona_id'].unique().to_list():
    list_w.append(f"'{word}'")
list_for_in = ', '.join(list_w)

query = f"""
SELECT *
FROM read_parquet('{path_data_persona}')
WHERE persona_id IN ({list_for_in})
"""
df_all_personas = pl.DataFrame(conn.execute(query).df())

# info of all personas profiles
list_w = []
for word in df_experiences_by_company_id['persona_id'].unique().to_list():
    list_w.append(f"'{word}'")
list_for_in = ', '.join(list_w)

query = f"""
SELECT *
FROM read_parquet('{path_data_profiles}')
WHERE persona_id IN ({list_for_in})
"""
df_all_personas_profile = pl.DataFrame(conn.execute(query).df())
df_all_personas_profile_f = df_all_personas_profile.group_by('persona_id').agg(pl.col('url').unique())

# info of all personas email
list_w = []
for word in df_experiences_by_company_id['persona_id'].unique().to_list():
    list_w.append(f"'{word}'")
list_for_in = ', '.join(list_w)

query = f"""
SELECT *
FROM read_parquet('{path_data_emails}')
WHERE persona_id IN ({list_for_in}) AND type == 'personal'
"""
df_all_personas_emails = pl.DataFrame(conn.execute(query).df())

def def_polars_fix_gmail(x):
    if "@gmail" in x:
        first_part = x.split('@')[0]
        second_part = x.split('@')[1]
        return f"{first_part.replace(".",'')}@{second_part}"
    else:
        return x

df_all_personas_emails_f = (df_all_personas_emails
                            .with_columns(pl.col('address')
                                          .map_elements(def_polars_fix_gmail, return_dtype=pl.String)
                                          .alias('normalised_emails'))
                            .unique('normalised_emails', keep='first')
                            .sort('persona_id')
                            .drop('normalised_emails')
                         )
df_all_personas_emails_f = (df_all_personas_emails_f.group_by('persona_id').agg(pl.col('address').unique(),pl.col('type').unique()))
df_all_personas_plus = df_all_personas.join(df_all_personas_emails_f, on='persona_id', how='left')

df_full_personas_who_worked_in_company = (df_personas_who_worked_in_company
                                       .join(df_all_personas_plus, on='persona_id', how='left')
                                       .join(df_all_personas_profile_f, on='persona_id', how='left')
                                      )

df_full_personas_who_worked_in_company = (
    df_full_personas_who_worked_in_company.with_columns(
        (pl.col("start_date").fill_null(pl.col("start_date").min()))
        .dt.year()
        .alias("start_year"),
        (pl.col("end_date").dt.year()).alias("end_year"),
    )
)

work_years = []
for i in range(len(df_full_personas_who_worked_in_company)):
    start_y = df_full_personas_who_worked_in_company[i, "start_year"]
    if df_full_personas_who_worked_in_company[i, "end_year"]:
        end_y = df_full_personas_who_worked_in_company[i, "end_year"]
    else:
        end_y = 2020

    tmp_work_years = []
    for y in range(start_y, end_y + 1):
        tmp_work_years.append(y)

    work_years.append(tmp_work_years)

df_full_personas_who_worked_in_company = (
    df_full_personas_who_worked_in_company.with_columns(
        pl.Series("work_years", work_years)
    )
)

# add hireups
title_name_match = ["ceo","chief","founder","owner","president","vp","vice","director",
    "cfo","cto","partner","head of","hr ","human","talent","senior","manager","lead"]

df_full_personas_who_worked_in_company = (df_full_personas_who_worked_in_company
    .with_columns(
        pl.when(pl.col('title_name').str.contains_any(title_name_match)).then(True).otherwise(False).alias("higher_up")
    ))



df_tmp_email_checker = (
    df_full_personas_who_worked_in_company
    .filter(
            pl.col('still_associated')==True,
            pl.col('address').list.len()>0
    )
        ['persona_id','address']
        .explode('address')
)

# if current_company_parquet.exists():
#     df_pre_full_personas_who_worked_in_company = pl.read_parquet(current_company_parquet)
#     list_pre_deliverable_address = df_pre_full_personas_who_worked_in_company['address'].drop_nulls().explode().to_list()
# else:
#     list_pre_deliverable_address = []

# list_of_emails_to_check = df_tmp_email_checker['address'].drop_nulls().to_list()
# list_lists_email_check = []

# var_total_emails = len(list_of_emails_to_check)
# var_current_email_count = 0

# def def_check_and_populate(email_to_check):
#     global list_lists_email_check, var_current_email_count
#     if email_to_check in list_pre_deliverable_address:
#         list_lists_email_check.append([email_to_check, True])
#     elif '@gmail' in email_to_check:
#         list_lists_email_check.append([email_to_check, True])
#     else:
#         try:
#             is_deliverable = check_deliverability(email_to_check)
#             list_lists_email_check.append([email_to_check, is_deliverable])
#         except:
#             list_lists_email_check.append([email_to_check, False])
#     var_current_email_count += 1
#     print(' '*10, end='\r')
#     print(round(var_current_email_count/var_total_emails,5), end='\r')

# with ThreadPoolExecutor(max_workers=20) as executor:
#     results = list(executor.map(def_check_and_populate, list_of_emails_to_check))

# df_email_check = pl.DataFrame(list_lists_email_check, schema=["address", "deliverable"], orient="row")
# try:
#     df_tmp_email_checker_f = (
#         df_tmp_email_checker
#             .join(df_email_check, on='address')
#             .filter(pl.col('deliverable')==True)
#             .group_by('persona_id').agg(pl.col('address').unique().alias("deliverable_address"))
#     )
# except:
#     df_tmp_email_checker_f = pl.DataFrame()

# if df_tmp_email_checker_f.is_empty():
#     df_full_personas_who_worked_in_company = df_full_personas_who_worked_in_company.join(df_tmp_email_checker.rename({'address':'deliverable_address'}), on="persona_id", how='left')
# else:
#     df_full_personas_who_worked_in_company = df_full_personas_who_worked_in_company.join(df_tmp_email_checker_f, on="persona_id", how='left')

# Info about personas experiences

# info of all experiences[]
list_w = []
for word in df_experiences_by_company_id['persona_id'].unique().to_list():
    list_w.append(f"'{word}'")

list_for_in = ', '.join(list_w)
query = f"""
SELECT *
FROM read_parquet('{path_data_experience}')
WHERE persona_id IN ({list_for_in})
"""
df_all_personas_experiences = pl.DataFrame(conn.execute(query).df())


# info of all comapnies in said experiences
list_w = []
for word in df_all_personas_experiences['company_id'].unique().to_list():
    if "'" not in word:
        list_w.append(f"'{word}'")

list_for_in = ', '.join(list_w)
query = f"""
SELECT company_id, company_name, company_industry, company_linkedin_url, company_location_country
FROM read_parquet('{path_data_companies}')
WHERE company_id IN ({list_for_in})
"""
df_all_companies = pl.DataFrame(conn.execute(query).df())

df_full_personas_experiences_plus = df_all_personas_experiences.join(df_all_companies, on='company_id', how='left')

df_full_personas_experiences_plus = (
    df_full_personas_experiences_plus
    .with_columns(
        pl.when(
            pl.col('company_id')==current_company_id
        )
        .then(True)
        .otherwise(False)
        .alias('target')
    )
)

# Info about personas education

# info of all experiences
list_w = []
for word in df_experiences_by_company_id['persona_id'].unique().to_list():
    list_w.append(f"'{word}'")

list_for_in = ', '.join(list_w)
query = f"""
SELECT *
FROM read_parquet('{path_data_education}')
WHERE persona_id IN ({list_for_in})
"""
df_all_personas_education = pl.DataFrame(conn.execute(query).df())


#ifon of allcomapnies in said experiences
list_w = []
for word in df_all_personas_education['school_id'].unique().to_list():
    if "'" not in word:
        list_w.append(f"'{word}'")

if list_w:
    list_for_in = ', '.join(list_w)
    query = f"""
    SELECT school_id, school_name, school_type, school_website, school_location_country
    FROM read_parquet('{path_data_school}')
    WHERE school_id IN ({list_for_in})
    """
    df_all_school = pl.DataFrame(conn.execute(query).df())
    
    df_full_personas_education_plus = df_all_personas_education.join(df_all_school, on='school_id', how='left')
else:
    df_full_personas_education_plus = df_all_personas_education

1 About the project

The project came to life after realizing that web scraping doesn’t allow deep-level filtering—without consuming too much time.The irony is, this project itself took me about a month, but the final RDB contains more data than I could ever scrape.

The raw data was 1.4 TB in size and holds information previously scraped.
Processing was done on my local machine using Python, Polars, and DuckDB, following this workflow:
- Processed raw data into structured Parquet files using Polars.
- Transformed each Parquet file into mini RDBs using Polars.
- Merged all mini RDBs into one using DuckDB.
- Analyzed and filtered data to fit the current project.

2 EDA

2.1 information technology and services indestry’s yearly new recruit count

list_of_unique_company_experience_years = []
for y in df_full_personas_who_worked_in_company['start_year'].unique().drop_nulls().to_list():
    if y not in list_of_unique_company_experience_years:
        list_of_unique_company_experience_years.append(y)
for y in df_full_personas_who_worked_in_company['end_year'].unique().drop_nulls().to_list():
    if y not in list_of_unique_company_experience_years:
        list_of_unique_company_experience_years.append(y)

list_year = []
list_state = []
list_count = []
list_names = []

def def_get_names_breked(tmp):
    if tmp.is_empty():
        names_string = ''
    else:
        tmp_list_name = []
        names_limit = 3
        row_limit = names_limit * 6
        for i, name in enumerate(tmp['full_name'].to_list()):
            ii = i+1
            tmp_list_name.append(name.title())
            if ii!=0 and ii%names_limit==0:
                tmp_list_name.append("<br>")
            if ii==row_limit:
                tmp_list_name.append("...")
                break
        names_string = ', '.join(tmp_list_name).replace(", <br>, ","<br>")
    return names_string

for y in list_of_unique_company_experience_years:
    #recuite state
    list_year.append(y)
    list_state.append('Recruited')
    tmp = df_full_personas_who_worked_in_company.filter(pl.col('start_year')==y).sort('full_name')
    list_count.append(len(tmp))
    list_names.append(def_get_names_breked(tmp))
    
    #recuite state
    list_year.append(y)
    list_state.append('Resigned')
    tmp = df_full_personas_who_worked_in_company.filter(pl.col('end_year')==y).sort('full_name')
    list_count.append(len(tmp))
    list_names.append(def_get_names_breked(tmp))

df_m_recruite_vs_resign = pl.DataFrame({
    'year':list_year,
    'status':list_state,
    'count':list_count,
    'names':list_names,})

2.2 mu sigma inc.’s workforce status over the years

3 Persona company network graph

gr_net = df_full_personas_experiences_plus.with_columns(pl.col('company_id').str.to_uppercase()).group_by('persona_id','company_id').agg(pl.len().alias('count')).sort('count')
list_top_in_network = gr_net['company_id'].value_counts().sort('count', descending=True)['company_id'].to_list()[:5]
gr_net_f = gr_net.filter(pl.col('company_id').is_in(list_top_in_network))

list_letters = ['A','B','C','D','E','F','G','H']
dict_company = {}
dict_company_rev = {}
for company, letter in zip(list_top_in_network, list_letters ):
    dict_company[letter] = company
    dict_company_rev[company] = letter

gr_gr_net_f = gr_net_f.sort('company_id').group_by('persona_id').agg(pl.col('company_id').unique().sort(),)

gr_gr_net_f2 = (
    gr_gr_net_f['company_id']
    .value_counts()
    .with_columns(
        # pl.col('company_id').list.join(', '),
        (pl.col('count')/len(gr_gr_net_f)).alias('per')
    )
    .sort('per',descending=True)
)

list_prob = []
for i in range(len(gr_gr_net_f2)):
    tmp_prob_letters = []
    for k in dict_company.keys():
        if dict_company[k] in gr_gr_net_f2[i]['company_id'][0].to_list():
            tmp_prob_letters.append(f' {k}')
        else:
            tmp_prob_letters.append(f'¬{k}')

    list_prob.append(f"P({' ∩ '.join(tmp_prob_letters)}) = {round(gr_gr_net_f2[i]['per'][0],4)}")
annon_prob_text = "<b>Probability Distribution:</b><br>" + '<br>'.join(list_prob)



# Create network graph
G = nx.Graph()
for persona, company in gr_net_f.select(['persona_id', 'company_id']).iter_rows():
    G.add_edge(persona, company)

# Get unique values
persona_ids = gr_net_f['persona_id'].unique().to_list()
company_ids = gr_net_f['company_id'].unique().to_list()

# Calculate degrees (connection counts)
degree_dict = dict(G.degree())

# Get min and max degrees for scaling
company_degrees = [degree_dict[c] for c in company_ids]
persona_degrees = [degree_dict[p] for p in persona_ids]

min_company_degree = min(company_degrees) if company_degrees else 1
max_company_degree = max(company_degrees) if company_degrees else 1
min_persona_degree = min(persona_degrees) if persona_degrees else 1
max_persona_degree = max(persona_degrees) if persona_degrees else 1

# Define size ranges
COMPANY_MIN_SIZE = 25
COMPANY_MAX_SIZE = 100
PERSONA_MIN_SIZE = 5
PERSONA_MAX_SIZE = 20

# print(f"Company connections range: {min_company_degree} - {max_company_degree}")
# print(f"Persona connections range: {min_persona_degree} - {max_persona_degree}")

# Sort companies by degree (size) in descending order
company_ids_sorted = sorted(company_ids, key=lambda x: degree_dict[x], reverse=True)

# Check if "Nokia" exists in the data
HIGHLIGHTED_COMPANY = current_company_id
HIGHLIGHTED_COMPANY_EXISTS = HIGHLIGHTED_COMPANY.lower() in [str(c).lower() for c in company_ids]

if HIGHLIGHTED_COMPANY_EXISTS:
    # Get the actual case-sensitive name
    highlighted_company = next(c for c in company_ids if str(c).lower() == HIGHLIGHTED_COMPANY.lower())
    # print(f"Highlighting company: {highlighted_company} (with {degree_dict[highlighted_company]} connections)")
else:
    # print(f"Warning: '{HIGHLIGHTED_COMPANY}' not found in company list")
    highlighted_company = None

# Create layout (companies on outer circle, ordered by size)
pos = {}
num_companies = len(company_ids_sorted)
radius_outer = 2.0

# Position companies on circle, ordered by size (largest first)
for i, company in enumerate(company_ids_sorted):
    # Start at top (90° or π/2 radians) and go counter-clockwise (add angle)
    # Counter-clockwise rotation: angle = start_angle + (i * 2π / num_companies)
    # This puts largest at top, next on left, then bottom, then right
    start_angle = np.pi / 2  # 90° at top
    
    # For counter-clockwise rotation
    angle = start_angle - (2 * np.pi * i / num_companies)
    
    # Convert to x, y coordinates
    pos[company] = (radius_outer * np.cos(angle), radius_outer * np.sin(angle))

# Position personas
for i, persona in enumerate(persona_ids):
    connected_companies = [c for c in company_ids if G.has_edge(persona, c)]
    if connected_companies:
        avg_x = np.mean([pos[c][0] for c in connected_companies])
        avg_y = np.mean([pos[c][1] for c in connected_companies])
        # Add jitter to spread out personas
        jitter_x = np.random.uniform(-0.2, 0.2)
        jitter_y = np.random.uniform(-0.2, 0.2)
        pos[persona] = (avg_x * 0.5 + jitter_x, avg_y * 0.5 + jitter_y)
    else:
        pos[persona] = (0, 0)

# Prepare edge traces
edge_x, edge_y = [], []
for edge in G.edges():
    x0, y0 = pos[edge[0]]
    x1, y1 = pos[edge[1]]
    edge_x.extend([x0, x1, None])
    edge_y.extend([y0, y1, None])

edge_trace = go.Scatter(
    x=edge_x, y=edge_y,
    line=dict(width=0.6, color='rgba(120, 120, 120, 0.15)'),
    hoverinfo='none',
    mode='lines')

# Prepare node traces with proportional sizing
company_x, company_y, company_text = [], [], []
company_color, company_size, company_hover = [], [], []
company_border_width = []  # For border thickness
company_border_color = []  # For border color

persona_x, persona_y = [], []
persona_color, persona_size, persona_hover = [], [], []

# Helper function to scale size proportionally
def scale_size(value, min_val, max_val, min_size, max_size):
    if max_val == min_val:
        return (min_size + max_size) / 2
    return min_size + (value - min_val) / (max_val - min_val) * (max_size - min_size)

# Add COMPANY nodes in sorted order (largest first)
for company in company_ids_sorted:
    x, y = pos[company]
    company_x.append(x)
    company_y.append(y)
    company_text.append(str(company))
    company_color.append('#EF553B')
    
    connections = degree_dict[company]
    # Scale size based on connection count
    scaled_size = scale_size(
        connections, 
        min_company_degree, 
        max_company_degree,
        COMPANY_MIN_SIZE, 
        COMPANY_MAX_SIZE
    )
    company_size.append(scaled_size)
    
    # Custom border for highlighted company
    if highlighted_company and company == highlighted_company:
        company_border_width.append(4)  # Thicker border
        company_border_color.append('#000000')  # Black border
    else:
        company_border_width.append(1)
        company_border_color.append('#000000')
    
    # Hover text
    personas = gr_net_f.filter(pl.col('company_id') == company)['persona_id'].to_list()
    rank = company_ids_sorted.index(company) + 1
    hover_text = f"<b>Company #{rank}:</b> {company}<br>"
    hover_text += f"<b>Personas worked here:</b> {connections}<br>"
    hover_text += f"<b>Connection rank:</b> {rank}/{len(company_ids_sorted)}<br>"
    if connections > 0:
        for persona in personas[:5]:
            persona_name = df_all_personas.filter(pl.col('persona_id')==persona)['full_name'][0].title()
            hover_text += f" • {persona_name}<br>"
        if connections > 5:
            hover_text += f" • ... and {connections - 5} more"
    company_hover.append(hover_text)

# Add PERSONA nodes
for persona in persona_ids:
    x, y = pos[persona]
    persona_x.append(x)
    persona_y.append(y)
    persona_color.append('#636efa')
    
    connections = degree_dict[persona]
    # Scale size based on connection count
    scaled_size = scale_size(
        connections,
        min_persona_degree,
        max_persona_degree,
        PERSONA_MIN_SIZE,
        PERSONA_MAX_SIZE
    )
    persona_size.append(scaled_size)
    
    # Hover text
    companies = gr_net_f.filter(pl.col('persona_id') == persona)['company_id'].to_list()
    persona_name = df_all_personas.filter(pl.col('persona_id')==persona)['full_name'][0].title()
    hover_text = f"<b>Persona:</b> {persona_name}<br>"
    hover_text += f"<b>Companies worked at:</b> {connections}<br>"
    if connections > 0:
        # Check if worked at highlighted company
        if highlighted_company:
            worked_at_highlighted = highlighted_company in companies
            if worked_at_highlighted:
                hover_text += f"<b>Worked at {highlighted_company}:</b> ✓<br>"
        
        hover_text += "<br>".join([f"  • {comp}" for comp in companies[:5]])
        if connections > 5:
            hover_text += f"<br>  • ... and {connections - 5} more"
    persona_hover.append(hover_text)

# Create company node trace
company_trace = go.Scatter(
    x=company_x, y=company_y,
    mode='markers+text',
    hoverinfo='text',
    hovertext=company_hover,
    text=company_text,
    textposition="top center",
    textfont=dict(size=14, color='black'),
    marker=dict(
        color=company_color,
        size=company_size,
        line=dict(
            width=company_border_width,
            color=company_border_color
        ),
        opacity=0.9)
)

# Create persona node trace
persona_trace = go.Scatter(
    x=persona_x, y=persona_y,
    mode='markers',
    hoverinfo='text',
    hovertext=persona_hover,
    text=None,  # No text for personas
    marker=dict(
        color=persona_color,
        size=persona_size,
        line=dict(width=1, color='black'),
        opacity=0.7)
)

# Calculate axis ranges for 1:1 aspect ratio
all_positions = list(pos.values())
x_vals = [p[0] for p in all_positions]
y_vals = [p[1] for p in all_positions]

# Add padding
x_range = [min(x_vals) - 0.5, max(x_vals) + 0.5]
y_range = [min(y_vals) - 0.5, max(y_vals) + 0.5]

# Make axes have the same range for 1:1 aspect
max_range = max(x_range[1] - x_range[0], y_range[1] - y_range[0])
x_center = (x_range[0] + x_range[1]) / 2
y_center = (y_range[0] + y_range[1]) / 2

x_range = [x_center - max_range/2, x_center + max_range/2]
y_range = [y_center - max_range/2, y_center + max_range/2]

# Create figure with 1:1 aspect ratio
fig = go.Figure(data=[edge_trace, persona_trace, company_trace],
                layout=go.Layout(
                    title=f'Persona-Company Network (Companies Ordered by Size)<br><sup>Highlighted: {highlighted_company if highlighted_company else "None"}</sup>',
                    showlegend=False,
                    hovermode='closest',
                    margin=dict(b=20, l=20, r=20, t=100),
                    xaxis=dict(
                        showgrid=False, 
                        zeroline=False, 
                        showticklabels=False,
                        range=x_range,
                        scaleanchor="y",
                        scaleratio=1
                    ),
                    yaxis=dict(
                        showgrid=False, 
                        zeroline=False, 
                        showticklabels=False,
                        range=y_range
                    ),
                    plot_bgcolor='white',
                    paper_bgcolor='white',
                    width=900,
                    height=900
                ))

# Add legend with size examples and highlighting info
# legend_text = f"""
# <b>Node Size = Connection Count</b><br>
# <span style='color:#EF553B'>● Companies</span><br>
# <span style='color:#636efa'>● Personas</span> (hover for details)
# """

# fig.add_annotation(
#     x=0.98, y=0.98,
#     xref="paper", yref="paper",
#     text=legend_text,
#     showarrow=False,
#     font=dict(size=14),
#     align="left",
#     bgcolor="rgba(255, 255, 255, 0.95)",
    
# )

# Add top companies list
top_companies = company_ids_sorted[:10]  # Top 10 companies
top_companies_text = "<b>Top Companies by Connections:</b><br>"
for i, company in enumerate(top_companies, 1):
    connections = degree_dict[company]
    top_connections = degree_dict[top_companies[0]]
    connections_per = f" | {round(connections/top_connections*100)}%" if highlighted_company and company != highlighted_company else ""
    highlight_indicator = " " if highlighted_company and company == highlighted_company else ""
    top_companies_text += f"{dict_company_rev[company]}. {company}: {connections} {connections_per} {highlight_indicator}<br>"

fig.add_annotation(
    x=0.02, y=0.98,
    xref="paper", yref="paper",
    text=top_companies_text,
    showarrow=False,
    font=dict(size=14),
    align="left",
    bgcolor="rgba(255, 255, 255, 0.9)",
    # bordercolor="#666",
    # borderwidth=1
)

# Add probabiliy list

fig.add_annotation(
    x=0.98, y=0.98,
    xref="paper", yref="paper",
    text=annon_prob_text,
    showarrow=False,
    font=dict(
        family="'Courier New', monospace",  # Multiple fallbacks
        size=12,
        color="black"
    ),
    align="left",
    bgcolor="rgba(255, 255, 255, 0.95)",
    
)
fig.write_image((path_output_images/f'network_{current_company_id}.webp'))
fig.show()

amount = 5

tmp = df_full_personas_who_worked_in_company.sort(
    ["inferred_salary", "linkedin_connections", "inferred_years_experience"],
    descending=True,
)
tmp_gr = df_full_personas_experiences_plus.group_by('persona_id').agg(pl.len().alias('experience_count'))
tmp = df_full_personas_who_worked_in_company.join(tmp_gr, on='persona_id').sort('experience_count',descending=True)

tmp2 = pl.concat(
    [tmp.filter(pl.col('still_associated')==True, pl.col('higher_up')==True)[:amount*2],
     tmp.filter(pl.col('still_associated')==True, pl.col('higher_up')==False)[:amount*2],
     tmp.filter(pl.col('still_associated')==False, pl.col('higher_up')==True)[:amount*1],
     tmp.filter(pl.col('still_associated')==False, pl.col('higher_up')==False)[:amount*1],
    ]
).sort("full_name")

list_persona_for_plot = tmp2['persona_id'].to_list()

# Workforce data

def def_plotly_experience_range(current_persona_id):
    tmp_df = (df_full_personas_who_worked_in_company
              .filter(pl.col('persona_id')==current_persona_id)
              .with_columns(pl.col('end_year').fill_null(2021))['start_year','end_year'])
    
    fig_tmp = copy.deepcopy(fig_company_hiring_trend)
    fig_tmp.add_vrect(
        x0=tmp_df[0,'start_year'],
        x1=tmp_df[0,'end_year'],
        fillcolor="blue",
        opacity=0.1,
        line_width=0 
    )
    return fig_tmp

def def_plotly_experience_gantt(current_persona_id):
    px_data = (df_full_personas_experiences_plus
               .filter(pl.col('persona_id')==current_persona_id)
               .with_columns(
                   pl.col('end_date').fill_null(datetime.datetime(2020, 1, 1, 0,0)),
                   pl.col('company_name').str.to_uppercase(),
                   # pl.col('company_name').str.to_uppercase().str.replace_all('&', '-and-')
               )
               .sort('start_date'))
    
    y_order = px_data['company_name'].to_list()
    
    fig = px.timeline(px_data,x_start="start_date", x_end="end_date", y="company_name",
                      color='target',hover_data=["title_name"], height=140+30*len(px_data),
                      category_orders={"company_name": y_order},
                      color_discrete_map={True:'#EF553B',  False:'#636efa'},
                      labels={'target':'Target', 'start_date':'Recruited', 'end_date':'If-Resigned', 
                             'company_name':'Company', 'title_name':'Job role'}
                     # title=f"Experience of {current_persona_name}.",
                     )
    fig.update_yaxes(
        # autorange="reversed",
                              showgrid=True,
                              gridcolor='lightgray',
                              gridwidth=1,
                              griddash='dot'
    )
    fig.update_layout(showlegend=False, xaxis_title=None, yaxis_title=None)
    return fig

4 Workforce sample

4.1 Abhijith Asok

Job title: Intern, analytical research
Socials: https://linkedin.com/in/abhijith-asok-50314330 | https://facebook.com/abhijith.asok.31 | https://twitter.com/abhijithasok

4.1.1 Abhijith Asok’s working period at mu sigma inc.

4.1.2 Gantt plot of Abhijith Asok’s experience

---

4.2 Anish Mahapatra

Job title: Trainee decision scientist
Socials: https://linkedin.com/in/anish-mahapatra-11b807101

4.2.1 Anish Mahapatra’s working period at mu sigma inc.

4.2.2 Gantt plot of Anish Mahapatra’s experience

---

4.3 Archisman Dinda

Job title: Trainee decision scientist
Socials: https://linkedin.com/in/archisman-dinda-a6b673b6 | https://twitter.com/archisman_dinda | https://github.com/archismandinda | https://linkedin.com/in/archismandinda | https://facebook.com/archismandinda

4.3.1 Archisman Dinda’s working period at mu sigma inc.

4.3.2 Gantt plot of Archisman Dinda’s experience

---

4.4 Ashish Kumar

Job title: Account manager
Socials: https://quora.com/ashish-kumar-140 | https://linkedin.com/in/ashish-kumar-5905a963 | https://linkedin.com/in/ashish-kumar-66a70b59 | https://linkedin.com/in/ashish-kumar-6b890811 | https://linkedin.com/in/ashish02 | https://facebook.com/ashishkumar01

4.4.1 Ashish Kumar’s working period at mu sigma inc.

4.4.2 Gantt plot of Ashish Kumar’s experience

---

4.5 Ashwin Avasarala

Job title: Engagement manager
Socials: https://linkedin.com/in/ashwin-avasarala-07ab841b | https://facebook.com/ashwin.avasarala

4.5.1 Ashwin Avasarala’s working period at mu sigma inc.

4.5.2 Gantt plot of Ashwin Avasarala’s experience

---

4.6 Aviral Agrawal

Job title: Trainee decision scientist
Socials: https://linkedin.com/in/aviral-agrawal-1172a496 | https://facebook.com/aviral.agrawal | https://linkedin.com/in/avirala

4.6.1 Aviral Agrawal’s working period at mu sigma inc.

4.6.2 Gantt plot of Aviral Agrawal’s experience

---

4.7 Balaji Soundararajan

Job title: Engagement manager
Socials: https://linkedin.com/in/balaji-soundararajan-mba-4408192 | https://linkedin.com/in/balajisoundararajan

4.7.1 Balaji Soundararajan’s working period at mu sigma inc.

4.7.2 Gantt plot of Balaji Soundararajan’s experience

---

4.8 Bryce Doreen

Job title: Senior consultant
Socials: https://linkedin.com/in/bryce-doreen-23aa334b

4.8.1 Bryce Doreen’s working period at mu sigma inc.

4.8.2 Gantt plot of Bryce Doreen’s experience

---

4.9 Chinedu Ndukwe

Job title: Engagement manager - data analytics, marketing and sales strategy
Socials: https://linkedin.com/in/chinedu-ndukwe-mba-889b057 | https://linkedin.com/in/chinedundukwe

4.9.1 Chinedu Ndukwe’s working period at mu sigma inc.

4.9.2 Gantt plot of Chinedu Ndukwe’s experience

---

4.10 Isha Vyas

Job title: Associate
Socials: https://linkedin.com/in/isha-vyas-a10a3952 | https://linkedin.com/in/ishavyas

4.10.1 Isha Vyas’s working period at mu sigma inc.

4.10.2 Gantt plot of Isha Vyas’s experience

---

4.11 Jayadev Chakravarty

Job title: Leadership team - japan
Socials: https://linkedin.com/in/jayadev-chakravarty-a2b148 | https://linkedin.com/in/jayadevc | https://angel.co/jchak | https://twitter.com/jchakravarty

4.11.1 Jayadev Chakravarty’s working period at mu sigma inc.

4.11.2 Gantt plot of Jayadev Chakravarty’s experience

---

4.12 Julie Woods-Moss

Job title: Advisor
Socials: https://linkedin.com/in/julie-woods-moss-ab76284 | https://twitter.com/juliewoodsmoss | https://linkedin.com/in/juliewoodsmoss

4.12.1 Julie Woods-Moss’s working period at mu sigma inc.

4.12.2 Gantt plot of Julie Woods-Moss’s experience

---

4.13 Mark Dzialga

Job title: Board of directors
Socials: https://linkedin.com/in/mark-dzialga-109893172

4.13.1 Mark Dzialga’s working period at mu sigma inc.

4.13.2 Gantt plot of Mark Dzialga’s experience

---

4.14 Neeratyoy Mallik

Job title: Team lead
Socials: https://linkedin.com/in/neeratyoy-mallik-85a74b87 | https://linkedin.com/in/nyoym

4.14.1 Neeratyoy Mallik’s working period at mu sigma inc.

4.14.2 Gantt plot of Neeratyoy Mallik’s experience

---

4.15 Noreen Madden

Job title: Business unit analyst
Socials: https://linkedin.com/in/nmaddenling

4.15.1 Noreen Madden’s working period at mu sigma inc.

4.15.2 Gantt plot of Noreen Madden’s experience

---

4.16 Pankhuri Vijay

Job title: Data analyst
Socials: https://linkedin.com/in/pankhuri-vijay-2ab717106

4.16.1 Pankhuri Vijay’s working period at mu sigma inc.

4.16.2 Gantt plot of Pankhuri Vijay’s experience

---

4.17 Piyush Mundhra

Job title: Leadership team, head of client services, u.s
Socials: https://linkedin.com/in/piyush-mundhra-88190ba | https://linkedin.com/in/piyushmundhra

4.17.1 Piyush Mundhra’s working period at mu sigma inc.

4.17.2 Gantt plot of Piyush Mundhra’s experience

---

4.18 Praveen Suresh

Job title: Apprentice leader- marketing
Socials: https://twitter.com/_praveensuresh | https://angel.co/praveen-suresh | https://facebook.com/praveen.farry | https://linkedin.com/in/praveensuresh

4.18.1 Praveen Suresh’s working period at mu sigma inc.

4.18.2 Gantt plot of Praveen Suresh’s experience

---

4.19 Raghavendra Ramesh

Job title: Business analyst | data science
Socials: https://linkedin.com/in/raghavendra-ramesh-b2b16b67 | https://linkedin.com/in/raghavendraramesh4 | https://twitter.com/raghuramesh4

4.19.1 Raghavendra Ramesh’s working period at mu sigma inc.

4.19.2 Gantt plot of Raghavendra Ramesh’s experience

---

4.20 Rahul Kishore

Job title: Trainee decision scientist
Socials: https://facebook.com/rahulkishore24 | https://linkedin.com/in/rahulkishore24

4.20.1 Rahul Kishore’s working period at mu sigma inc.

4.20.2 Gantt plot of Rahul Kishore’s experience

---

4.21 Rajat Sharma

Job title: Regional head
Socials: https://linkedin.com/in/rajat-sharma-02b5b96 | https://linkedin.com/in/rajat-sharma-3a545b1b | https://linkedin.com/in/rajat-sharma-62355b4 | https://linkedin.com/in/rajat-sharma-71a47536

4.21.1 Rajat Sharma’s working period at mu sigma inc.

4.21.2 Gantt plot of Rajat Sharma’s experience

---

4.22 Ramon Serrallonga

Job title: Data science manager
Socials: https://twitter.com/r_serrallonga | https://linkedin.com/in/ramonserrallonga

4.22.1 Ramon Serrallonga’s working period at mu sigma inc.

4.22.2 Gantt plot of Ramon Serrallonga’s experience

---

4.23 Rituparna Roy

Job title: Trainee decision scientist
Socials: https://linkedin.com/in/rituparna-roy-90362885

4.23.1 Rituparna Roy’s working period at mu sigma inc.

4.23.2 Gantt plot of Rituparna Roy’s experience

---

4.24 Saul Bravo

Job title: Junior associate
Socials: https://linkedin.com/in/saulcornejobravo

4.24.1 Saul Bravo’s working period at mu sigma inc.

4.24.2 Gantt plot of Saul Bravo’s experience

---

4.25 Sreekanth Mahesala

Job title: Senior associate
Socials: https://linkedin.com/in/sreekanth-mahesala-b50b8445 | https://facebook.com/sreekanth.mc | https://github.com/sreekanthmahesala | https://linkedin.com/in/sreekanthmahesala

4.25.1 Sreekanth Mahesala’s working period at mu sigma inc.

4.25.2 Gantt plot of Sreekanth Mahesala’s experience

---

4.26 Stephen Wu

Job title: Business analytics consultant
Socials: https://indeed.com/r/stephen-wu | https://linkedin.com/in/stephen-wu-02885b32 | https://linkedin.com/in/stephenwu81618

4.26.1 Stephen Wu’s working period at mu sigma inc.

4.26.2 Gantt plot of Stephen Wu’s experience

---

4.27 Sujit Menon

Job title: Leadership - sales
Socials: https://linkedin.com/in/sujit-menon-80b63b4 | https://linkedin.com/in/sujitmenon

4.27.1 Sujit Menon’s working period at mu sigma inc.

4.27.2 Gantt plot of Sujit Menon’s experience

---

4.28 Vinayakam Murugan

Job title: India leadership team
Socials: https://linkedin.com/in/mvinayakam

4.28.1 Vinayakam Murugan’s working period at mu sigma inc.

4.28.2 Gantt plot of Vinayakam Murugan’s experience

---

4.29 Vivek Jha

Job title: Data scientist - associate
Socials: https://linkedin.com/in/vivek-jha-48730069 | https://linkedin.com/in/vivekjha2 | https://linkedin.com/in/vivekjhausa

4.29.1 Vivek Jha’s working period at mu sigma inc.

4.29.2 Gantt plot of Vivek Jha’s experience

---

4.30 Yuedong Lu

Job title: Key accounts manager
Socials: https://linkedin.com/in/yuedonglu | https://twitter.com/yuedonglu

4.30.1 Yuedong Lu’s working period at mu sigma inc.

4.30.2 Gantt plot of Yuedong Lu’s experience

---

df_full_personas_who_worked_in_company.write_parquet(current_company_parquet)